Executive brief
Ollama, a platform for running large language models locally, contains a security vulnerability in its model downloading component. An attacker could potentially trick the server into making unauthorized requests to internal or external systems. This could lead to the exposure of sensitive internal information or allow the attacker to interact with other services on the local network that are not intended to be public.
Technical details
A Server-Side Request Forgery (SSRF) vulnerability exists in Ollama versions up to 0.18.1 within the Model Pull API. The flaw is located in the 'server/download.go' file, where improper validation of user-supplied input allows for request manipulation. A remote attacker with low privileges can exploit this to force the server to initiate network requests to arbitrary destinations. This can be used to scan internal networks, access local services, or bypass network-based access controls. As of the disclosure date, the vendor has not responded to the report.
Affected products
- Ollama Ollama up to 0.18.1
Timeline
- 2026-04-05: disclosed: Initial public disclosure via VulDB/NVD